Intel details Clearwater Forest Xeon built on 18A with new Darkmont e-cores

Intel unveiled Clearwater Forest, a next-generation E-core Xeon, at Hot Chips, pairing major core architecture changes with the 18A process and advanced 3D packaging.

Intel used Hot Chips to present Clearwater Forest, its next-generation E-core Xeon, positioning the design as a showcase for its foundry and packaging capabilities. The chip replaces Sierra Forest and is built around the new Darkmont efficiency cores. Intel said Clearwater Forest will ship in a large chiplet package that uses the 18A node for compute dies and relies on Foveros Direct 3D stacking combined with EMIB for advanced multi-die integration.

At the core level, Darkmont emphasizes a wider and more parallel front end and a strengthened out-of-order engine. Intel disclosed a 64 KB instruction cache per core and described an expanded decoder that can handle more instructions per cycle than the previous Crestmont E-core design. The company also detailed enhancements to reorder and allocation structures, an increase in allocation units, and an enlarged out-of-order window to support more in-flight work.

Execution resources were significantly increased, with many more execution ports for integer and vector math to sustain much higher parallel throughput. Intel framed the combination of architectural upgrades, the 18A process, and advanced 3D Foveros packaging paired with EMIB as a demonstration of what its foundry and packaging teams can now deliver. Specific shipping timelines, power or performance numbers, and final system-level configurations were not stated.

65

Impact Score

AMD and Rackspace plan dedicated AI compute rollout

AMD and Rackspace have finalized a phased deployment for dedicated AMD-based compute across Rackspace data centers. The capacity is aimed at regulated enterprise workloads, including clinical AI and large-scale inference.

Lexar tests SSD offloading for local AI models

Lexar is developing an AI-focused SSD approach designed to cut DRAM demand when running large language models on consumer PCs. Internal tests show the company’s storage offloading can load models that traditional local frameworks struggle to run with limited memory.

NVIDIA Blackwell leads MLPerf Training 6.0

NVIDIA’s latest MLPerf Training 6.0 results put Blackwell across every benchmark in the suite, including new MoE workloads. Partner systems from Microsoft Azure and CoreWeave highlighted large-cluster runs on Llama 3.1 405B and DeepSeek-V3 671B.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.